Machine Learning Estimates of G20 Subnational GHG Emissions 2000-2020

This preprint announces the development of a machine learning framework to estimate annual Scope 1 and 2 CO2-equivalent emissions for subnational jurisdictions in G20 countries from 2000 to 2020.

• The model integrates geospatial, socioeconomic, and environmental data with self-reported emissions inventories, aligned with subnational administrative boundaries, improving spatial relevance and predictive accuracy (R2=0.77, MAPE=38.57%).
• The dataset covers 5,972 cities and 116 regions in G20 countries, leveraging multiple data sources and advanced AutoML techniques (AutoGluon), and aims to provide a globally consistent baseline for assessing subnational climate progress, especially where data are scarce or inconsistent.